ORIGINAL PAPER International Journal of Occupational Medicine and Environmental Health 2016;29(1):15 – 39 http://dx.doi.org/10.13075/ijomeh.1896.00509 FACTORIZATION METHODS APPLIED TO CHARACTERIZE THE SOURCES OF VOLATILE ORGANIC COMPOUNDS IN MONTREAL, QUEBEC EUGENIUSZ PORADA1 and TERMEH KOUSHA2 1 University of Québec at Outaouais, Gatineau, Canada Department of Computer Science 2 Ottawa University, Ottawa, Canada Department of Mathematics and Statistics Abstract Objectives: The study objective was to assemble emission characteristics of the sources of the ambient volatile organic com- pounds (VOCs) and to elaborate methods of organizing them into the sources’ chemical profiles. Material and Methods: The UNMIX – sensor modeling method from the U.S. Environment Protection Agency (EPA) – was used to process the VOC concentration data acquired over the years 2000–2009 for 175 VOC species in 4 air quality monitoring stations in Montreal, Quebec. Results: The method enabled to assess VOC emissions from the typically distributed sources exist- ing in urban environment and VOC occurrences characterizing the local, or point-like, sources. The distributed sources were inextricably associated with hydrocarbons from exhaust, heavier hydrocarbons from contaminated urban soil, fugi- tive evaporations of gasoline and liquefied petroleum gases (LPG), leakage from the industrial and commercial use of solvents, and the inert, ozone depleting gases permeating urban atmosphere. The sources’ profiles were charted involv- ing 60–120 VOC species per source. Spatial distribution of the sources was examined. Conclusions: The UNMIX applica- tion and the source profiling methods, by building robust chemical profiles of VOC sources, provided information that can be used to assign the measured VOC emissions to physical sources. This, in turn, provides means of assessing the impact of environmental policies, on one hand, and of industrial activities on the other hand on VOC air pollution. Key words: Volatile organic compound, UNMIX, Receptor modeling, Source profile, Pulp industry, Principal components INTRODUCTION due to the extent of exposure in the densely populated Man-made volatile organic compounds (VOCs) con- areas [1,4–10]. Atmospheric photochemistry converts tribute significantly to the urban air pollution. Widely the more reactive VOCs by oxygenation and through occurring VOCs, such as e.g., benzene, ethylbenzene, reactions with oxides of nitrogen to secondary VOCs, xylenes, 1,1,1-trichloroethane, trichloroethylene, tetra- free radicals, ozone, and subsequently to smog organic chloroethylene, carbontetrachloride, chloroform, are to- aerosols [6,11,12]. xic and/or carcinogenic substances [1–3]. In large cities, Correlating VOC emissions from anthropogenic sourc- harmful effects of increasing ambient concentrations of es with public health indices allows to assess the im- pollutants are even more dangerous to the public health pact of VOC pollution on human health. The research Received: November 14, 2014. Accepted: January 24, 2015. Corresponding author: E. Porada, 1081 Alenmede Cr., Ottawa ON, K2B 8H2, Canada (e-mail: [email protected]). Nofer Institute of Occupational Medicine, Łódź, Poland 15 ORIGINAL PAPER E. PORADA AND T. KOUSHA provides guidelines for planning emission control strat- MATERIAL AND METHODS egies. Achieving reduction in VOC concentrations in VOC measurements and data the ambient air comparable to the reduction of the “clas- As part of the National Air Pollution Surveillance (NAPS) sic” pollutants (from 1970 to 2008 particulate mat- federal program in Canada [31], air quality monitor- ter concentration in the ambient air has decreased by ing stations were installed across the country, several of over 50% [13,14]) seems feasible. them in Montreal, Quebec. The sensors operate basing Source apportionment of numerous VOC species have on gas chromatography/flame ionization detector sys- already been examined [15–24]. The research presents tem (GC/FID) for quantification of light hydrocarbons in the methods of modeling the ambient VOC multi-re- the C2 fraction, while a combined gas chromatography/ ceptors operating in the air quality monitoring stations. mass selective detector (GC/MSD) system operating in The chief methods applied are the Principal Compo- selected ion monitoring (SIM) mode is used for quantifi- nent methods – mainly the Positive Matrix Factoriza- cation of heavier hydrocarbons, in the C3 to C12 fractions. tion (PMF). Also Chemical Mass Balance (CMB) system The data used for this report consists of daily aver- provides methods of modeling the receptors [25,26]. ages of VOC species concentrations in the ambient air. The methods consist in constructing sources’ chemical Measurements of 175 species were taken each 6th day fingerprints – employing a menu of VOCs – in a form of in 4 monitoring stations located in the Great Montreal sources’ emission profiles, relying on the user-specified area (Figure 1). The concentrations are expressed in mi- inventory of the sources contributing to the measured crograms per cubic meter. totals. Stations 2 (No. 50104) and 3 (No. 50115) are located in The intricacies of organic air pollution and complex at- downtown areas, close to a large body of water, one of mospheric chemistry make it challenging to build a reli- them, i.e., station 2, is adjacent to a large multi-industrial able receptor model [26–29]. The VOC menu typically park. Station 1 (No. 50103) is located in the East Montreal lists the most prominent VOCs, with high signal/noise NAPS stations [No.] ratio (SNR), which allow a clear-cut profiling of the sour- 1: 50103 2: 50104 N ces. Thereby, some point sources, particularly the speciali- 3: 50115 1 4: 50129 zed industrial plants and the associated waste streams, 3.5 km which yield quite distinct chemical fingerprints of low-con- centration VOCs, may stay unresolved. 2 4 This research addresses the issue by developing robust 3 pulp and paper source discrimination routines, which also search for industry the low-emission sources that stay out in the usual hy- industrial park stations drocarbon background in the urban air. This is achieved by applying principal component (PC) methods imple- Montreal-East (station 1) – petrochemical industry district; light grey mented by the U.S. Environmental Protection Agen- (stations 2, 3) – densely populated urban area; dark grey – green terrains, suburbs, and semi-rural districts in the inner ring (station 4); cy (EPA): EPA UNMIX [30], which has the advantage white – body of water surrounding inner-ring Downtown Montreal. over the other PC methods of telling the user which Fig. 1. Demarcation of districts and localization of the National modeling attempt cannot be successful, thus, narrowing Air Pollution Surveillance (NAPS) monitoring stations the output to the reliable receptor models. in Montreal, Quebec 16 IJOMEH 2016;29(1) SOURCES OF VOCs IN MONTREAL ORIGINAL PAPER zone, formerly a borough of Montreal, a concentration of not apply the “positive matrix” constraint; the emission the petrochemical and chemical industries in the Great profiles allow for negative values. Persistent negative val- Montreal area. Station 4 (No. 50129) is located on the pe- ues may signal VOC depletion in the atmosphere. ripheries of Montreal, in a more countryside-like region, Finally, UNMIX possesses an important feature, which but crossed by highways. In the neighborhood there is enables to make robust informed guesses regard- also a pharmaceutical industry. ing VOC sources: it fails to give results when the par ti- This report concerns VOC measurements taken over cular profiling cannot be made reliable. The success of the years 2000–2009, during the period when effects profiling depends, in an essential way, on the collec- of implementation of several international and federal tion of VOC species fed into the routine. By combining clean-air programs [14,32,33] were already clearly visible. all the results from the successful run – thus, combining The number of all the measured VOC species in each only partial reliable profiles defined in different limit- station was 175, but many measurements bore errors. In ed VOC ranges – more reliable exhaustive chemical de- this research, only the species that have 95% or more cor- scriptions of VOC sources can be produced. rect daily averages in the study period were considered. Therefore, the number of VOCs adequately measured The method of combining partial emission profiles and processed varies between 103 and 130, depending on into the sources’ extended profiles the monitor. Modeling the 175-VOC multi-receptor with UNMIX re- quired applying the routine several hundred times in order U.S. EPA VOC data processing routines to produce sufficiently exhaustive library of the reliable The United States EPA makes 2 data processing ap- (according to UNMIX), but only partial (approximate- plications based on the Principal Component method, ly 30 VOCs could be fed at a time into the routine), emis- i.e., EPA PMF and EPA UNMIX available. The methods sion profiles. Thus, 500 to 700 (depending on the moni- perform linear modeling of many variables (VOC mea- tor) profiles were created to ensure that each of the 100 surements) with a few expressly constructed predictors or to 130 correctly measured VOCs participated in a success- factors – which are alleged apportionments of the mea- ful profiling. A successful model featured 5 to 8 sources, sured emissions among putative sources. The applications depending on the VOC menu